Research On CSI-based Behavior Recognition For Indoor Scenario Identification

Location based services(LBS)is in a trend of development,people used outdoor LBS to recommend local news and delicious restaurants in the past,at present the demand for LBS has gradually expanded to indoors,people pay more attention to whether it is able to recommend better taste restaurant in the mall in real time,whether it is able to introduce the cultural relics in front of you in the museum in real time.All of these require complete indoor maps combined with indoor positioning,however indoor positioning has developed to a relatively mature stage now,many methods could achieve decimeter-level indoor positioning through the triangular positioning or fingerprint positioning.In the field of indoor map construction,the indoor maps constructed in most of the current research are often “lattices” without room types,and how to mark the real scenes represented by these “lattices” in the indoor map is the research object of this paper.Most existing indoor scene recognition methods use crowdsourcing to collect user mobile phone data and combine different types of sensors data such as camera,microphone,Wi-Fi,and accelerometer to determine the scene type.The existing methods generally have some limitations,for example,in most methods,private data such as camera data and microphone data are utilized,which is easy to invade user privacy on the one hand,and cannot be recognized in a scene where image collection is not allowed on the other hand.In addition,the power consumption of the device is expensive and the crowdsourcing data of dozens or even hundreds of people is often required in the acquisition process to achieve the recognition of a scene.In order to solve the shortcomings of the above researches,this paper proposes an indoor scene recognition method only based on the existing WiFi infrastructure.Extracting the parsed CSI through Wi-Fi to achieve fine-grained environment awareness including behavior recognition and crowd counting.In this paper,the channel state information(CSI)fluctuations are fully analyzed,and the indoor scenes are classified by the percentage of nonzero elements(PEM)algorithm.The behavior recognition method of zero prior knowledge is proposed to achieve the behavioral awareness of indoor scenes,and then a behavior based decision tree classifier is designed to determine the type of scenes.This paper evaluates the feasibility of the method in the 84 rooms of the experimental building and the shopping mall.The results show that the method achieves an accuracy of about 90.5%.